Engine × version that produced the numbers (empty for older data)
benchmarks
Table
Per-engine, per-benchmark scores (repeats collapsed to their median)
scores
Table
Engine × benchmark scores (ordered by total_score)
runtimes
Table
Engine × benchmark times in seconds (ordered by total)
computeScores(runs)
Function
Compute scores table from filtered runs
computeRuntimes(runs)
Function
Compute runtimes table from filtered runs
aq
Global
Arquero library
op
Global
Arquero operators
Scoring: Repeated runs of the same
engine and benchmark are first collapsed to their median time — timings are
right-skewed, since a run can be slowed by unrelated load but never made faster than the
machine allows. For each benchmark, the geometric mean of those times across all engines is
computed. An engine's score for a benchmark is gmean / time (faster = higher).
Total score is the sum across benchmarks. The runs column records how many
repeats each row summarizes.